 In recent years, deep neural network-based transfer learning, TL, has been successfully applied to EEG-based motor imagery, MIF, BRAINTH, computer interfaces, BCIs, but its application remains limited due to the difficulty of selecting appropriate source domains and the lack of model independence. To address these issues, this paper proposes a multi-direction transfer learning, MDTL, strategy, which uses multiple source domains to improve the target domain's performance. Additionally, the proposed approach is model independent, allowing it to be easily deployed on existing models. The results show that MDTL outperforms traditional TL methods, reducing the training time by up to 93.94%. Furthermore, MDTL can be used with any existing deep learning models, making it a versatile and reliable tool for EEG-based MIBCI applications. This article was authored by Ong Li, Zhenyu Wang, Xijiao, and others. We are article.tv, links in the description below.